The training phase converges quickly (acc>0.95), but the validate result is very bad (acc<0.3)

[UESTC-Liuxin]

您好，我用我自己的数据集（汉英，真实场景160k数据量）进行实验，发现训练很快收敛，但是验证结果很差，您出现过这种情况吗？我想的话，这是不是因为这种结构和输入方式，相当于设置了teaching_forcing = 1，很容易就导致过拟合了。

[charlesmindee]

Hi, I have the same issue

[jiangxiluning]

@UESTC-Liuxin 你好我没有出现过具体问题，会不会跟你文字长度有关吶

[charlesmindee]

Actually I had made a mistake in the loss, I fixed it shifting the ground-truth sequences to the right! (I changed a bit the loss function in my implementation, now the model is working well when predicting)

[jiangxiluning]

@charlesmindee great!